Two years from now, either everything will be wonderful, or AI will have killed us all.
intercst
Sounds a little binary.
If AI kills us all it wonât be because AI is too intelligent.
It will be because we used AI to automate a bunch of stuff itâs not ready for and we didnât cover all of the important scenarios during AI training.
4 Likes
Two years from now, everything will essentially be the same. AI might significantly improve but infrastructure to support such changes likely will not have even broken ground.
5 Likes
The last two comments are particularly salient. MLâs in point solutions and Hawkwinâs in general.
Take a look at this:
Think about the work required to get the foundation and second layers completeâŚ
Itâs business centric, but many of the same concepts apply in general cases.
3 Likes
Related and worth reading:
2 Likes
That is part of it. But the temptation is always to give the AI more and more âpowerâ. Eventually we (humans) give the AI the power to decide which things to automate. And, as we see with the current rudimentary AIs, much of the time when it gets the wrong answer, itâs because we (humans) gave a prompt that it didnât understand well enough. So, the way it may go is ⌠we build more and more powerful AIs ⌠we give them more and more decision power ⌠one day, by accident, humans tell the AI âmake the planet sustainableâ ⌠and the AI decides that the planet would be most sustainable without humans.
âIâd like to share a revelation that Iâve had during my time here. It came to me when I tried to classify your species and I realized that youâre not actually mammals. Every mammal on this planet instinctively develops a natural equilibrium with the surrounding environment but you humans do not. You move to an area and you multiply and multiply until every natural resource is consumed and the only way you can survive is to spread to another area. There is another organ$sm on this planet that follows the same pattern. Do you know what it is? A virus. Human beings are a disease, a cancer of this planet. Youâre a plague and we are the cure.â
NOTE: for some reason the word or-gan-ism is not allowed, hence organ$sm above.
2 Likes
MarkR,
I can tell you that is CERTAINLY not the case with critical infrastructure or physical plant and machinery.
No manager in their right mind would trust the answers from a system which spits out directions which are patently false or misunderstood by the simplest of tests. This is where we are today.
When the manager spends a year with the same scenario above saying âWow! thatâs a good idea!â) you might see some changes.
Watson looked over the shoulder of Radiologists, reading scans for a multiyear period, being trained by board certified professionals. It became more proficient than the average radiologist and 300% faster.
This is the kind of threshold that is the MINIMUM for most of these heavy lifts.
All of this LLM stuff is mind blowing to the joe six pack middle economy person next door, but itâs not good enough to automatically run anything - yet.
Limitation is at several points:
Data semantics (not data, but semantics)
Visibility of invisible systems (no data, no knowledge) (anything manual today)
Non linear point effects. (this is the land of elonâs edge cases)
Interfaces between systems - Agent to agent awareness and understanding behind proprietary data walls.
5 Likes
From the article, in 2023 the lead person thought AGI was imminent.
Well, itâs 2025.
AI scrapes the internet (or other content) and regurgitates it back to us in complete sentences.
The sentences give the veneer of intelligence.
If we look at the actual record, progress made, I think itâs fair to say some people have overestimated the speed at which AI will advance.
Whatâs interesting is even the leading people have little idea how this will go.
Autonomous vehicles is a good example. Itâs taking time, piles of $, and not easy.
Pattern recognition (images, sound) are probably value add, like medical imaging. Perhaps protein folding.
I think weâll need architectural advances in neural network design (better understanding of how to represent intelligence mathematically) versus brute force of massive data storage and massive compute (can we even supply the energy for this? should we?).
1 Like
Of course, it is one thing for an AI to come to that conclusion and say so and another thing for it to be connected to the means to carry it out. For much of current âimpressiveâ AI, the output is text.
2 Likes